ISPOR Annual Conference 2026

May 17, 2026

Evidence Prime contributed four research posters to ISPOR 2026, held in Philadelphia from 17 to 20 May. Artur Nowak represented the team on site, sharing work on deduplication, screening and structured extraction with the health economics and outcomes research community.

Four questions from our research

  • MSR26: prospective follow-up of SuperDeduper examined deduplication across three systematic reviews, including how confidence levels can direct uncertain duplicate groups to human review. The focus was on evaluating the complete decision process, including records that should not be removed automatically.
  • MSR77: adaptive screening thresholds and abstention tested a strategy that can include, exclude or postpone a screening decision. The simulations examined both ordinary screening errors and more difficult failure conditions. They also exposed an important limitation: agreement between a person and AI can conceal missed studies when their errors are correlated.
  • MSR129: suspected-drug identification in pharmacovigilance explored structured extraction with supporting quotations, document locations and rationales. This study addressed one component of identifying a valid safety case. Extracting every required case element remained a subsequent research step.
  • MSR179: fine-grained patient-flow extraction examined numbers assessed for eligibility, randomised and lost to follow-up, preserving study-level and study-arm context. The work combined structured outputs with source evidence and qualitative review of extraction errors.

The posters brought together contributions from Artur Nowak, Ewelina Sadowska, Joanna Konieczna, Ewa Borowiack, Monika Opalek, Iwona Kmicikiewicz and Damian Stachura. The linked abstracts provide the author list and study details for each presentation.

Evaluation that reflects real evidence work

In his first-day conference recap, Artur reflected on the gap between encouraging benchmark results and performance in a particular review. The discussions reinforced the need to examine ambiguous records, difficult cases, abstention and human verification alongside average performance. His second-day recap applied the same questions to pharmacovigilance and structured extraction.

Across these projects, the practical question was how to give evidence teams outputs they can inspect and challenge. These were evaluations of defined tasks and workflows; their findings need to be understood within those study settings when considering use in a different review.

Blank white image with no visible content or details